General > General Technical Chat

Injection locking the 10Mhz OCXO to external reference (upgrading a FY6600)

(1/6) > >>

Johnny B Good:
 I'm looking for a means to add an external 10MHz reference clock input to my much upgraded FY6600 AWG that eliminates the use of a change over switch and disruption to the internal 50MHz clock generated by the 10MHz OCXO I used to upgrade the previous  50MHz 0.1ppm TCXO upgrade to the rather crappy SMD XO chip originally fitted.

 I'd originally contemplated just adding an SMA - F socket and a small change-over switch to the rear panel and hope for the best which, if the need arose, I could add some complex switching logic to minimise disruption to the 50MHz clock. However, a reference to "injection locking" in relation to a Wien Bridge oscillator gave me inspiration to delve a little deeper into the whole question of such synchronisation of oscillators. It looks an effective way to achieve my goal but duckduckgo searches failed to turn up any inspiration as to how best to "injection lock" an actual OCXO, hence this post.

 I can think of two possible ways to 'inject' the external GPSDO 10MHz reference into the OCXO. The first would be to use a small transformer to add the minimal level of 10MHz ripple to the 12v Vdd line required to drag it into lock by the 100 to 500ppt offset range with the second being application of the 10MHz reference as a ripple voltage to the Vfc control pin.

 Despite the presence of any internal decoupling components on these lines in the OCXO, given enough 'brute force and ignorance' and the susceptibility of any XO to lock onto an external signal less than 0.5ppb off frequency, I think there's every chance of this tactic working. If it does, it'll be a more elegant alternative to the socket and change-over switch method implemented by Arthur Dent in the FY6600 thread over a year ago.

 Has anyone else used this method of implementing an external 10MHz reference socket upgrade on comms and test gear? Any advice regarding pitfalls and such or suggestions to offer before I start experimenting?  :)

 In the meantime, you can gaze at a couple of image files I've attached. The collection of 10MHz OCXOs was kicked off by the 13MHz example I snagged at a recent radioham rally for just 4 quid. I used it to drive a 2 times clock multiplier to drive a divide by 13 (74193) which then drove a 5 times clock multiplier which gave me an ultra low jitter 10MHz square wave. However, in my search for a data sheet for this 5 volt part, I chanced upon a uk based ebay trader offering those 10MHz 12 volt OCXOs for just a penny shy of a fiver each. I bought three just over a week ago then another four a few days later, hence the collection. One is fitted to the AWG and, at just a fiver each, one can never have too many 10MHz OCXOs.  :)

 Now that I have a "Lifetime's Supply", I can disclose my source. He's still got "more than ten available"  >:D  I must point out that the description is less than honest in that they're ex-equipment rather than NOS (you could even see the odd through plated hole attached to a pin or two) but, at that price, one can hardly complain since they're otherwise in very good condition with plenty of tuning control range left in them.

<https://www.ebay.co.uk/itm/CQE-CRYSTAL-OSCILLATOR-10MHz-REDUCED-TO-CLEAR/253081992039?hash=item3aecdcbb67:g:x1gAAOSw~qNZh2rl>

<https://tinyurl.com/y6pl2xvk>

 The second image is my attempt at showing the effect of a couple of 10MHz Xtals on the rather jittery and saw toothed 10MHz square wave output on the PPS line of my u-blox M8N based arduino.Rpi GPS module. For a still image, it's a pretty fair representation but what you can't see is the rather toned down sawtooth correction jitter (4 per second to one every few seconds - even tens of seconds). Although not perfect, it's a surprisingly huge improvement over the raw PPS signal I'd previously been using as a frequency calibration source.

JBG

Johnny B Good:
 Hello all,

 I noticed that the photos did attract some interest since I posted them almost a fortnight back, despite the absence of any suggested solutions to my question being offered. Since some interest seems to have been aroused by this thread, I thought I'd report on the latest developments by way of "closure".  :)

 Several days after my initial post, I decided to do some simple tests to determine if it was even possible to 'injection lock' these CQE OCXOs and how best to achieve this goal. Sadly for me during the later phase of my tests, I managed to blow the PPS line on my precious NEO M8N based GPS module with a jolt of 12 volts. The sorry tale can be read here:-

https://www.eevblog.com/forum/projects/ublox-neo-m8n-gps-navigation-signal-amplify-module-for-arduino-rasppery-pi/msg2426871/#msg2426871

 That's not the end of the story since I bought a cheap 4 quid NEO-6M to continue my experiments. It doesn't do "10MHz" on its PPS line, just a maximum of 1KHz but this, surprisingly, still suffices as a calibration reference by which to adjust the OCXO fitted into my FY6600 function generator (and of course, can still be used to discipline an OCXO - the frequency of the PPS is pretty well immaterial to a DIY GPSDO anyway).

 This time however, I kept the GPS module well out of harms way (as I should have done with the original unit  :palm:) and set two of those OCXOs up on the breadboard as a master/slave setup to continue my aborted test program. I tried injecting the output from the "Master" into the "Slave's" Vfc pin with no joy, then into the Vref pin before trying the output pin (I didn't bother trying to inject into the 12v Vcc pin) which finally did show the desired response.

 Either the oscillator output isn't buffered at all or, (more probably) simply not very well buffered since I was able to drag the slave's frequency some +&-17ppb or more by carefully tuning the 'master' (which would likewise be dragged off frequency by the slave - the effect being mutual in this case) before 'the elastic broke' and the waveform degenerated into an amplitude modulated mix of the two OCXO outputs.

 I didn't bother to lash up an amp to buffer the master OCXO from the 'pulling effect' of the slave, which I'd expect to double the pulling range, since even a mere +/-10ppb will more than suffice to lock it to a GPSDO reference since, in this case,  I'd consider even a 1ppb error as being 'beyond the pale'.

 One thing I did notice, after removing the 100 ohm series resistor I'd originally installed from the master's output connection, was the complete nulling out when reconnecting to the slave's output pin before the signal stabilised. I need to lash up a buffer amp on the master's output to simulate the effect of a GPSDO reference source and use a Schmitt trigger buffer on the slave's output to stabilise the level being fed into the 3N502 multiply by 5 chip which has replaced the original 50MHz SMD XO so that I can feed the 'injection' signal into the slave's output pin at 3 to 6dB down to eliminate this total nulling out effect.

 Hopefully, even at some 3 to 6dB down, the master's buffered output in this test setup will still demonstrate the desired injection locking effect (and over at least a +/-10ppb range). I've obviously got a lot more testing to do but this does look a promising alternative to just simply using a change-over switch between the internal and external clock reference sources with all the pitfalls that such a scheme entails (unwanted breaks and switching transients).

 Assuming these tests verify my 'injection locking' technique, I'll have to take drive levels into account when designing the add on level limiting and buffering circuit (I think a single 74HC14 IC will suffice here) since I'm currently relying on the relatively high output impedance of the OCXO's 4Vpp sine wave output to save destructively over-driving the 3N502 multiplier which is powered off the 3.3v rail on the main board. I suspect the multiplier chip won't take so kindly to the direct output from a 74HC14 output powered from a 5 volt rail (precautions will need to be taken).

 I'll report on further progress as and when I actually make any. In the meantime, I've no further images to offer at this stage but, if a justification to include any arises later on, I can always do a brief follow up if I can't add them in a later edit.

JBG

[EDIT 2020-02-04]

 "Good news Everyone!"  :)

 After finally completing my Basic DIY GPSDO project a fortnight or so back, I was able to finally resume this frequency locking to an external 10MHz reference add on module project for my FY6600 (much to the seeming amusement of the minor deities, "Murphy" and "Sod" - at times it hadn't been so much a case of "Two steps forward, one step back." so much as more like "One step forward and two steps back."  ::) ).

 Anyway, suffice to say that I finally completed the project late this afternoon and, with the signal generator finally locked to my GPSDO 10MHz reference, I'm currently in the middle of a brief three hour experiment to determine whether or not the μHz frequency adjustment is a "Real Thing" or just a cosmetic feature of this dirt cheap signal generator.

 I've just realised in hindsight, that this is a test that could be performed without locking the sig genny to an external 10MHz reference. You set one channel to exactly 10MHz sine wave (or Sinc pulse) to use as the trigger reference and simply observe the second channel set to generate a Sinc pulse with a 1μHz offset dialled in.

 Sine based waveforms (and possibly including the triangle waveform) will average out the 4ns jitter inherent to the DDS technology used in this sig genny when not using the "magic" frequencies based on integer division of its 250MHz DAC sampling clock frequency (50, 25 and 12.5 MHz for examples - running with a 1μHz offset is never going to be any such "magic" frequency anyway - "Close, but no cigar." being the applicable phrase in this case). You can still see the effect of (in this case) the 4ns p2p jitter (see attached 'scope screen shots) but it's easier to assess the average mid point along the X axis than when using square wave forms. Note: the second image was captured just 5 hours after starting the test run when the peak had been centred just 0.5ns to the left of the trigger marker line.

 To be able to observe a relative drift of just 1ns (about the minimum required for an initial assessment) requires a run time of 2 hours and 47 minutes alone which puts this endeavour into the class of "Watching paint dry". A more complete experiment to observe a half cycle drift (50ns at 10MHz) requires a run time of 138 hours and 53 minutes ( The best part of a week!) which puts it in the category of "Watching paint peel."

 Some 2 1/2 hours in and it seems to have shifted by the anticipated 1ns but I discovered about an hour ago whilst tilting the sig genny to observe the gravitation effect on my injection locking module's attempts at holding the internal OCXO on frequency (and in phase) that the T adapter connection to the scope's CH2 BNC, which is monitoring the GPSDO 10MHz reference signal, was less than stellar in that it could introduce a variation of circa 1ns depending on the angle of dangle of the cables. However, an overnight run should answer the question of "Real or Cosmetic?".  :)

 I'll post a more detailed report in the next few days once I've gotten over my surprise at actually completing not just one project but two! The GPSDO project took me just over a year to complete and this frequency injection locking project just a mere eight months!  ::)

[END_EDIT]

Johnny B Good:
As promised in my latest (now 6 day old) edit of the previous posting I'd made on the 31st of May last year, I'm providing a detailed update on the culmination of this injection locking project.
 
Regarding the cheap NEO-6M I referenced in that May 2019 post, that proved to be a complete waste of 4 quid as far as my plans for building a simple PLL based GPSDO were concerned.

Programming the PPS line to output 50% duty cycle 1ms pulses so as to run the Phase Detector at a less glacial rate than 1Hz failed because it suffered from an entirely unexpected behaviour whereby at the halfway point in a 10 to 20 minute or so cycle time (which seemed dependant on how far from exactidue the on-board 48MHz XO happened to be), the nicely square shaped pulses would start slimming down until they finally disappeared up their own fundament for a split second before reincarnating as full fat 50% duty cycle pulses once more for another turn around the block.

I suffered almost a whole week's worth of frustration in trying to get my PLL to lock the OCXO to the GPS module's 1KHz "square" wave output before it finally occurred to me to monitor exactly what was being presented to the phase detector inputs. Even then, it didn't become immediately apparent until I'd kept a watchful eye on the 'scope traces for over ten minutes.

The problem proved intractable. My work around of using the class II phase detector in an ancient 4046 I'd been testing with and programming the narrowest possible 1μs wide pulse so as to restrict the funny business of it disappearing up its own fundament to the last split second of a 10 to 20 minute cycle failed to dilute the effect in a vast ocean of time sufficiently to avoid corrupting the PLL process.
 
Having tried a long shot at resolving this issue, I then had no choice but to order a cheap NEO-M8N module to replace the very peculiar NEO-6M I'd chanced 4 quid on before I could resume testing of my basic GPSDO project and transfer it from the solderless breadboard stage onto the vero-board inside of an actual project box stage. Thankfully, this 15 quid purchase arrived just a week later. It was a 'fake' (no flash - but a CR2032 coin cell overcomes the half hour data retention time limit). Fake or not, in this case it really didn't matter and I was able cobble a working GPSDO together on my breadboard test setup.
 
In the end, I used an even cheaper 8 quid 'fake' M8N to complete the GPSDO project since the 15 quid fake was around 10dB 'deafer' than the original genuine module I'd blown out the PPS line on and this second even cheaper fake was only some 2 or 3dB down on the original (I suspect this merely lacks the flash rather than the LNA and SAW filter components as well). My external GPS antenna has a direct view of a Cellphone tower that's only 103 metres away which can desense these GPS modules by some 2 to 6dB as evidenced by a whole constellation's worth of SV signals dropping some 2 to 6dB in almost complete unison (there's usually one or two SV's that defy this desense hypothesis) each time it fires up in my direction.

Since I'm now forced to rely on Ebay and Banggood for my component supply, this adds considerable delay to any such projects, hence the rather protracted hiatus in this injection locking project of mine. I needed a working GPSDO reference before it would make any sense to resume this project once more.

Having finally commissioned a MK I GPSDO some three weeks ago (I plan on making a MK II version later in the year), I was able to turn my thoughts to designing myself a frequency locking injection module for that aforementioned FY6600 sig genny.

Now that I had a 10MHz +12dBm sine wave into 50 ohm reference that couldn't be pulled off frequency by any loads, I could finally lash up a test circuit onto a solderless breadboard with one of the half dozen CQE 10MHz OCXOs I happen to have going spare. I did try injecting into the Vref and Vfc pins but found injecting into the output pin still gave the best result.

Curiosity over the impedance my injected signal was driving into led me to run a basic impedance test on all 6 of these CQE OCXOs which ranged from a low of 75 ohms through to a high of 90 ohms. A lot lower than I had expected but not the 50 ohm impedance so typically quoted for many sine wave output OCXOs. Until that point in time, it had never occurred to me to test the output impedance of these OCXOs - I'd just assumed it wouldn't really matter a whole lot in practice.

In all my (inexcusable) ignorance, I hadn't realised just how much of a beating my precious 3N502 clock multiplier chip had been taking when the sig genny was powered down with the OCXO left running from its own 12v psu kept live whilst the generator remained plugged into a live mains outlet. The injection locking module now has the honour of guarding my precious 3N502 (precious 'cos they cost something like a fiver a pop versus the ten for a quid 74HC14 chips) from this onslaught.

At least if the 74HC14 in the injection locking module succumbs to this abuse, I still have two unused inverters going spare as substitutes (if that ever happens I'll add a 100 ohm resistor in series to reduce the input protection clamp diodes' current to a less stressful level). In any case, I'd much prefer having to replace a 10 pence part than a 5 quid one.

My hunch that a single 74HC14 IC would suffice proved correct but I did add a BC547 buffer amp between the extl 10MHz reference input BNC socket and the single inverter I was using to drive the 10.4MHz 3 pole Butterworth LPF I was using to convert the square wave back into a reasonable facsimile of a sine wave at a well defined voltage level to inject into the OCXO's output pin which itself was driving another inverter biassed to its mid-point to generate a reliable square wave immune to the inevitable variations in signal level arising out of injecting the external reference at some 3 to 6dB down on the OCXO's output.

At this stage of the breadboard testing phase, I needed to know how much level of injection signal I could apply before the risk of completely nulling out the oscillator's output voltage (quite possibly completely stalling the oscillator which is even worse) and how reliably the inverter gate could convert it back to a well defined square wave voltage to drive the 3N502 multiplier I was using to recreate the original 50MHz clock from the 10MHz OCXO which had replaced the cheap and nasty 50MHz XO chip originally used in the Feeltech design.

Being on soldelress breadboard, there was significant 'ground bounce' both within the circuitry and on the 'scope probe connections to muddy the waters but I did persevere with my testing until I felt I was ready to commit it to a copper ground plane with the hex schmitt inverter mounted 'dead bug' style.

Knowing I'd have to reduce the level coming out of the LPF to eliminate the risk of completely killing the OCXO's output signal, I designed the filter for 300 ohm impedance and put a 300 ohm resistor in series between the output of the inverter and the filter's input with a 100 ohm resistor in  series in the feed to the OCXO's output pin.

This didn't work out quite as well as I'd hoped so, just on a hunch alone, I put a 10MHz crystal in series with the filter's output which improved the pulling range but I discovered the effect was very assymetric in that whilst I could drag the frequency down by over 30ppb, I could rarely pull it up by more than 5ppb with any of the six OCXOs I tried. I figured the ground bounce effect from the solderless breadboard construction was aggravating this problem so it wasn't until I started testing the final module built onto the copper groundplane that I finally figured out a solution.

Basically, I had overlooked the fact that the series resonant mode of any quartz crystal resonator that this simple filtering effect depended upon was just slightly lower than the marked frequency (perhaps just a KHz or two in this case), requiring the use a low value capacitor in series in order to shift Rs up to 10MHz exactly. I had a vague notion that something in the range of 15 to 30pF would be required so wired a 45pF compression trimmer in series with the crystal and placed this across the 10MHz feed to the 'scope and tuned the trimmer for minimum signal voltage.

Surprisingly, this test worked even better than just shorting out the half metre stub of croc clip ended BNC cable I was using for the test connection. Even more encouraging was the fact that the LC meter showed a value of 26.6pF which closely matched the 26.4pF reading from one of two or three 27pF caps I'd sampled. With the chosen 27pF cap in series (and the 100 ohm resistor now permanently removed), the assymetric effect was more than amply reduced for my purpose (a desired requirement to reliably pull the OCXO a minimum of +/-10ppb from its nominal frequency).

It has now occurred to me that a properly trimmed 10MHz quartz crystal resonator is probably all that's required to filter the inverter's output, making that 3 pole Butterworth LPF redundent (or at least replaceable with a simple RC LPF). However, by that stage, I hadn't wanted to risk upsetting what may have been a fortuitously arrived at solution and add more delay to the project.

I still have a spare 10MHz crystal and a bunch of OCXOs to experiment with at my leisure to see whether I can trim the fat off any MK II injection locking module designs. All this complexity might beg the question as to the need for such but as I've already mentioned, my concern is to avoid gross disruption to the signal generator's main clock supply which I fear could crash critical processes within the FPGA logic.

Since I need to limit the level of injection into the OCXO's output to avoid this situation and only inject another sine wave so the only distorting effect is that of amplitude ripple within limits that will allow the schmitt trigger inverter to reliably convert it to a 3v logic level without missing a single beat, I need to take a sine or square wave 10MHz reference input, convert it to a well defined 5v logic level square wave to drive a filter that outputs a reasonable facsimile of a likewise well defined level of sine wave to inject into the OCXO's output without risk of disruption to the feed to the inverter which drives the input of an attenuator network feeding the 3N502 on the main board with the required 3v logic level clock pulses, hence this "complicated" solution. My fears may be unfounded but my previous experience with a 50MHz TCXO module would suggest otherwise.

The copper groundplane mentioned above is actually an 8mm micro-bore copper pipe slit and flattened out into a 3 inch long 1 inch wide sheet which I bent into a tall L shape to act as the module's chassis held onto the back panel of the sig generator by the BNC connector. Whilst this offered a neat solution (groundplane and chassis mount all in one), it did prove a bit of a pig when it came to soldering the ground connections.

Indeed, it led to the ruination of two of the only three push-button switches that I had in my stock suited to this task which I had attempted to solder down to the 'chassis' for use as a term/thru option switch and some (fortunately repairable) damage to the last of these switches before I realised the plastic parts just couldn't tolerate soldering temperatures being applied to their metal bodies for more than a few seconds at a time. This was one example of "One step forward, two steps back." mentioned in my edit.

I did figure out a way round this problem, suffice to say that once I'd mounted the one surviving switch, I kept a short length of brass bar clamped underneath the switch location with a pair of molegrips for the rest of the build to limit the temperature rise whilst soldering the remaining groundplane component connections. This also provided a convenient means of holding it steady in a small vice clamped to the workbench for the remaining build time.

I'd initially considered mounting all of this extra circuitry onto the OCXO board itself until I realised the problems it would present to the switchable termination circuit I wanted to include to provide maximum flexibility in regard of my 10MHz lab reference feed options. Also, I was a little reluctant to pull the existing OCXO board apart for a major rebuild. Going the separate sychronising module route meant I merely had to intercept the existing OCXO coaxial feeder cable to the mainboard's original XO chip location (now occupied by the 3N502 multiplier chip).
 
I was simply going to chop into the existing co-ax feed between the OCXO module and the mainboard and solder a couple of SMA male plugs onto the cut ends which would plug into the SMA F sockets I'd elected to use on the synchronising module. I cut the cable but after all the faff of soldering a plug onto the cable coming off the mainboard, I decided it would be easier to solder an SMA socket onto the through plated hole perf board I'd built the OCXO module onto and use a handy 15cm straight to angled male to male SMA patch lead for this connection.

Remarkably, these modifications to the OCXO module had seemingly little to no effect on its calibration setting other than, inevitably for an entirely anticipated modest amount of retrace (less than half a ppb's worth at most).
 
A star feed distribution from a multi-channel amplifier is the ideal solution for such distribution but my GPSDO only sports a single output port requiring the addition of a distribution amplifier at a later date. Whilst the FY6600 is currently the only piece of lab kit with such an Extl 10MHz reference input right now, I didn't want to forego the through/term option's flexibility that would allow me to daisychain the 10MHz reference feed onto other kit if needed.

Even if I had a 4 way distribution amp and cables, it could still prove a handy feature when adding a fifth item of kit that could also benefit from such a reference source. If a job's worth doing, then (in general) it's worth doing properly as far as I'm concerned.

Before installing the module, I took the precaution of running a few tests which, contrary to some (well, a lot) of my previous projects, failed to reveal any show stopping issues from wiring errors and solder blob short circuits (Murphy must have been taking a break from his usual fun 'n' games activities that day). Testing the frequency bending limits showed that I still had good margins of stability to at least +12ppb and to -25ppb (the assymetry hadn't entirely disappeared but it was muted enough for my purpose).

Satisfied that it was as ready as it was going to be, I then mounted it into the sig genny which involves a level of commitment since I can't solder the 51 ohm terminating resistor and the 330 ohm 'stopper' resistor connection to the buffer amp input onto the BNC centre pin until after tightening up the BNC retaining nut which can only be done by turning the whole connector using a T adapter as a 'spanner' (there simply isn't any clearance for a spanner to spin the retaining nut - just barely enough room to wedge it with a pair of long needle nose pliars).

After plugging the mains cord back in and waiting a few minutes for the OCXO to reach operating temperature (its 12v psu mains connection isn't switched), I switched the sig genny on and was rewarded with a fully functioning generator that was still within 200ppt of calibration. Connecting the external 10MHz reference dropped its voltage to half and gently dragged the 10MHz Sinc pulse back onto frequency exactly as I'd anticipated.

This was all 'fine and dandy' but it was only 0.2ppb adrift to start with, not a very demanding test. I had to check out my injection locking module's behaviour at more extreme offsets which meant applying the trimtool to 'vandalise' my previous calibration efforts of three or more months back. I was able to maladjust by a margin that comfortably exceeded my minimum of +/-10ppb (IIRC, it was something like +15 to -25 ppb before things started looking decidedly ragged but without any sign of losing lock).

I've recalibrated the sig genny's OXCO to within +/- 0.2ppb but until now it would usually vary this much each day seemingly due to changes in room temperature of around 5 to 6 deg C pk2pk despite the XO being ovenised and wrapped in a sponge rubber overcoat to both shelter it from the modest cooling air flow from a small (40mm square by 10mm deep) 5 volted 12v axial fan and reduce the OCXO's internal temperature gradients (and energy consumption - a mere 850mW now). I guess once you're into ppt territory, this is where double ovening comes into its own.
 
However, at this level of precision, this is also where the effects of gravity start to loom large. Tilting the sig genny between face down and face up (180 deg pitch change) generates a delta T of 4ns per second which is a 4 ppb change in frequency when the internal OCXO is not locked to an external reference.

Even tilting the sig genny up on its bail stand (little more than a 15 deg upward tilt) is enough to show an immediate change so it's important to choose between level or tilted orientation before attempting any recalibration of the OCXO reference. I've made use of this 'feature' in the past to temporarily compensate for the diurnal temperature variations to avoid interfering with the hard won calibration setting of the frequency trim pot.

There's a possibility that this add on module might prove to be a buffer against changes in the loading on the OCXO's output from the clock multiplier chip arising out of these daily temperature changes in the main board but it's early days and up till now, I've been observing effects only visible when locked to a GPSDO reference. Once I become bored with observing the sig genny's 10,000,000.000000Hz Sinc pulse output running slow to the tune of 1.5ns per day (equates to a loss of one second every 1.8 million years!), I'll allow it to free run in order to observe whether this (hypothesised) ambient temperature variations effect has been mitigated to any extent.

Regarding my comments in the last edit of the previous posting in relation to the question of the utility of the 1μHz resolution settings, it does seem to be a real (if ever so slightly flawed) thing. Also, it turns out that you don't need to lock the generator to a precise and stable external reference in order to observe this 1ns drift per 167 minutes. Indeed, you don't even need to upgrade the original 20ppm rated XO to run the test which is best done by setting both channels to generate a Sinc pulse at exactly 10MHz using one as a trigger reference with the other offset by 1μHz.
 
When I ran this test, the horrible 4ns random jitter completely cancelled out and I was able to more accurately observe the drift rate. With a +1μHz setting it took slightly longer than the anticipated 17 minutes to drift by 0.1ns and with a-1μHz offset, it was more like 16 minutes. The 10MHz isn't one of this generator's magical frequencies so may well be straining the accuracy limits of its DDS. I haven't tried repeating these tests at any of the magical frequencies (12.5, 25 or 50MHz) as yet but I suspect this slight error may well vanish in these test cases (and likewise when comparing the 25 or 50MHz sine output against the 10MHz GPSDO reference).

 I'll run more tests as I find the time and report any further findings here at a (possibly much) later date. In the meantime, I've attached a picture of the hand drawn circuit diagram I'd been working from. It's obviously rather scruffy but, as far as I can see, an accurate representation (for the most part) of what I put together.

 BTW, the pinout sketch of  the 1117 was added when I'd been considering it as an add-on to the OCXO board to be fed from its 12v supply. It didn't make it onto the BoM once I'd decided it was going to be a separate add on module powered directly from the main board's 5v analogue supply.

 For the benefit of anyone curious about what it was resting upon, I can reveal that it was a Kenwood TS-140S HF Transceiver that happened to be a convenient support surface nicely exposed to even daylight illumination from my office/workshop window.
 
 JBG
 

Labrat101:
Hi ,
after reading this with all the time and research you put into it . Impressive. I spent , I forgot how many
hours, days, weeks . I spent getting my FY6800 cheapy . to work well. If any of us get these units to work
to Feeltech Data spec sheet which was comparable to some of the high end stuff on the market.
That's why I bought it in the first place . I just could not believe the specs over price.
I replaced the entire output op's power supply ,oxco , added heat sinks on power supplies ,etc.
There was one thing you said about the output of the oxco should be matched to the PCB you used 100ohm .. this in fact should be about 80 ohm to match the PCB's capacitance. 
I don't remember the exact details did it about 9 months ago and my memory at 67 is getting  :-// .
But I did notice but never tried the empty vref pins on the main chip . there is 4 references there and are not used I was guessing these are for injection for different levels .
The jitter I got it down to almost stable @ 20Mhz with every other square  :palm:
 But I think I will live with this as is ..
If I were to add time cost parts brain juice , coffee etc into the equation . I could Have build a Fusion Reactor  and had Free Electric .. :palm: (Murphy there is no such thing as Free anything ).
Keep Posting I really like your line of thought.

Johnny B Good:

--- Quote from: Labrat101 on May 07, 2020, 09:11:33 pm ---Hi ,
after reading this with all the time and research you put into it . Impressive. I spent , I forgot how many
hours, days, weeks . I spent getting my FY6800 cheapy . to work well. If any of us get these units to work
to Feeltech Data spec sheet which was comparable to some of the high end stuff on the market.
That's why I bought it in the first place . I just could not believe the specs over price.
I replaced the entire output op's power supply ,oxco , added heat sinks on power supplies ,etc.
There was one thing you said about the output of the oxco should be matched to the PCB you used 100ohm .. this in fact should be about 80 ohm to match the PCB's capacitance. 
I don't remember the exact details did it about 9 months ago and my memory at 67 is getting  :-// .
But I did notice but never tried the empty vref pins on the main chip . there is 4 references there and are not used I was guessing these are for injection for different levels .
The jitter I got it down to almost stable @ 20Mhz with every other square  :palm:
 But I think I will live with this as is ..
If I were to add time cost parts brain juice , coffee etc into the equation . I could Have build a Fusion Reactor  and had Free Electric .. :palm: (Murphy there is no such thing as Free anything ).
Keep Posting I really like your line of thought.

--- End quote ---

 Well, having recently 'celebrated' my 70th birthday, I can well appreciate your concerns over your memory. :( In this case, I think you may be confusing somebody else's contribution for mine (possibly an Arthur Dent posting, maybe?). I don't recall mentioning the need to use a 100 ohm matching resistor in my own OCXO modification. In fact, as I mentioned somewhere in this thread (see! my memory fails me even for stuff I wrote just a few weeks ago - serves me right for posting such overstuffed missives I suppose :palm:), the output impedance of my batch of OCXOs hadn't even crossed my mind until I started working on this injection locking project.

 I've since determined that they range from a low of 75 ohms to a high of 90 ohms. In this case, I needed to have some idea of what the output from my injection locking circuit would be facing in order not to overdrive the output from the OCXO. I merely want to coax it into locking to the external reference without risk of nulling it out or possibly stalling it completely (which would be even worse).

 I've deliberately avoided matching the OCXO's output impedance on the injection locking module's OCXO input specifically to facilitate the injection locking process - the short 15cm or so of RG316 is only being used as a screened feed rather than as a co-axial transmission line in this case and is short enough to present no impedance mismatch problems that would otherwise arise in metre's worth of co-axial cable at 10MHz.

 The co-ax link from the module's output feeding the clk input of the 3N502 clock multiplier is even shorter and although its clk input is a high impedance (compared to 50 ohm coax at any rate), I decided to create a 50 ohm attenuator network to drop the 5v TTL output to the 3.3v logic level required by the clock multiplier chip as a token gesture impedance matching exercise in honour of the use of 50 ohm coax as short a length as it was. At the very least, any reflections from the "far end" of this unterminated coaxial line will be swiftly damped.  :)

 As for the issue of the 4ns jitter on square waves in the FY6600 and 6800 models, that's not going away any time soon. You may be able to hide it from your 'scope traces by clever triggering schemes but, except for the "special magic frequencies" it's still there, nonetheless. The only reason why I even thought to try and verify the claimed μHz resolution was on account I now had sufficient frequency accuracy and stability courtesy of now being able to lock its reference to a GPSDO.

 It was only after some protracted experimentation that I finally stumbled upon the "Set both channels to the exact same frequency with a Sinc pulse waveform and offset one by just 1μHz" method and monitor their now completely jitter free 'scope traces regardless of any XO upgrade. This is a test any owner of an FY6600 or 6800, upgraded or not, can do if they have a 2CH DSO to hand (older CRO types might not have the nanosecond resolution or triggering accuracy of the more recent DSO offerings).

 It may only be of academic interest (you'd need a reference accurate and stable to one part in 10E14 for it to be any real use - even the best GPSDOs fall short of this by an order or two of magnitude  :palm:). The one thing you can say about this is that Feeltech aren't shortchanging you on frequency resolution. ::) Accuracy, yes but definitely not a lack of resolution even if you do lock it to an atomic clock based reference.  :popcorn:

 I do have more developments to report but I'll leave that for a later and more polished posting. I'm going to compose it in a text file to (hopefully) keep it rather more concise than my more usual postings of late.

JBG

Navigation

[0] Message Index

[#] Next page

There was an error while thanking
Thanking...
Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod